// 01 — TIERS
Three ways to buy compute.
// ON-DEMAND
On-demand
Per-GPU-hour. Launch and terminate at will.
Billed per GPU-second, region multiplier applied.
// RESERVED
Reserved
1 / 6 / 12-month commit. Lowest committed rate.
Up to 40% off on-demand for a 12-month commit.
// SPOT
Spot
Interruptible capacity, cheapest rate.
Up to 65% off. Preemptible with a 2-minute interruption notice.
// 02 — ON-DEMAND RATES
| GPU | On-demand | Reserved | Spot |
|---|---|---|---|
| H100 SXM | $2.49/hr | $1.49/hr | $0.87/hr |
| H200 SXM | $3.59/hr | $2.15/hr | $1.26/hr |
| B200 | $5.89/hr | $3.53/hr | $2.06/hr |
| B300 | $7.95/hr | $4.77/hr | $2.78/hr |
Rates are per GPU. Region multiplier (1.00×–1.22×) applied at metering time.
// 03 — RACK-SCALE
// PLATFORM SERVICES
Every service is metered per-second where applicable; the region multiplier applies at record time. Prices come straight from the metering catalog.
| SKU | Price |
|---|---|
| build-minutes | $0.008/minute |
| bandwidth-gb | $0.08/gb |
| edge-invocations | $0.6/million |
| seat | $20/seat-month |
Set budgets and hard spend caps in Console → Billing.
// FAQ
- How is compute billed?
- On-demand GPU compute is metered per second. Reserved commitments and interruptible spot trade flexibility for a lower rate. A region multiplier is applied at metering time.
- How do the platform services get priced?
- Every service meters into one ledger and prices come straight from the metering catalog, so the rates shown here match what you are billed. Most services are metered per second where applicable.
- Can I cap my spend?
- Yes. Set a per-org spend cap with a soft alert and a hard stop in Console → Billing. When the hard stop is reached, new launches are paused automatically.
- Does pricing change by region?
- The base rate is the same; a region multiplier is applied at the time usage is recorded to reflect regional cost.
// GET STARTED
Only pay for what you meter.
Per-second billing, a region multiplier applied at record time, and hard spend caps you control.